Predicting Power Flow and Wind Capacity Factor Using Integrated Spatio-Temporal Approach

This research proposes a novel spatio-temporal approach that integrates convolutional long short-term memory (ConvLSTM) networks and graph neural networks (GNNs) to model and predict wind power generation and its impact on power flow. The methodology uniquely combines ConvLSTM for capturing wind generation dynamics with GNN-based power flow analysis, offering a unified framework for renewable energy grid integration. Testing on IEEE standard systems (14-300 bus) demonstrates the approach's scalability and computational efficiency, achieving up to 11x faster computation compared to traditional Newton–Raphson methods. Applied to wind generation scenarios in the Norwegian grid, the ConvLSTM model achieves an R${}^2$ value of 0.977 in forecasting wind generation dynamics, while the GNN model demonstrates robust power flow prediction capabilities with an R${}^2$ of 0.948. This scenario-based framework bridges wind prediction and power flow analysis, enabling efficient grid performance assessment under varying wind conditions, while offering improved computational efficiency for real-time renewable energy integration and grid management.